Research On Multi-scale Set-valued Information System And Its Application

With the explosion of information,how to extract the useful knowledge from data is an important issue in current knowledge discovery research,especially in the case of the multi-granularity and multi-scale knowledge description.In a multi-scale information system,each attribute may have different attribute values according to different measurement scales,and there is information granule transformation process among different scales.It is known that the decision results obtained under different scales may be different,under which measurement scale the decision result is the best,that is,the optimal scale,then how to choose the optimal scale,is a hot issue in multi-scale research.At present,most of the researches on multi-scale information system consider that the value of object is unique at all scales of the attribute.However,in some practical problems,the value of some attributes exists in the form of set.In addition,most of the existing optimal scale selection models only focus on consistency or uncertainty,ignoring the cost in practical application.Therefore,in view of the above problems,this thesis has carried out research,the main content is as follows:Considering that the value of each attribute for each object is unique,this thesis introduces the concept of set-valued into the multi-scale information system,thus,we put forward the concept of multi-scale set-valued information system.Based on the semantic background of set values,the similarity relationship between objects is determined,and then the classical rough set theory is used to describe the approximate concepts and related properties in different scales.Then four kinds of constraint functions are proposed,which are upper approximation distribution,lower approximation distribution,positive region,and generalized decision in multi-scale set-valued decision system,and then the corresponding optimal scale selection is defined,and the relationships among different goals of optimal scales is discussed from the perspective of consistency and inconsistency.Considering that each object may have different scale level about some attributes in practical problems,the concept of generalized multi-scale set-valued information system based on the multi-scale set-valued information systems is proposed in this thesis.Aiming at the cost problem in practical application,a generalized multi-scale set-valued information system with cost-sensitivity is defined.By introducing the idea of sequential three-ways decision,a new method of updating scale space is proposed to improve the time efficiency.Then,the optimal scale selection model is established to minimize the uncertainty and cost with user demand as the constraint condition.Finally,the validity of the model is verified by experiments.